为什么Python Lambda有用？[关闭]

I’m trying to figure out Python lambdas. Is lambda one of those “interesting” language items that in real life should be forgotten?

I’m sure there are some edge cases where it might be needed, but given the obscurity of it, the potential of it being redefined in future releases (my assumption based on the various definitions of it) and the reduced coding clarity – should it be avoided?

This reminds me of overflowing (buffer overflow) of C types – pointing to the top variable and overloading to set the other field values. It feels like sort of a techie showmanship but maintenance coder nightmare.

Are you talking about lambda functions? Like

lambda x: x**2 + 2*x - 5

Those things are actually quite useful. Python supports a style of programming called functional programming where you can pass functions to other functions to do stuff. Example:

mult3 = filter(lambda x: x % 3 == 0, [1, 2, 3, 4, 5, 6, 7, 8, 9])

sets mult3 to [3, 6, 9], those elements of the original list that are multiples of 3. This is shorter (and, one could argue, clearer) than

def filterfunc(x):
    return x % 3 == 0
mult3 = filter(filterfunc, [1, 2, 3, 4, 5, 6, 7, 8, 9])

Of course, in this particular case, you could do the same thing as a list comprehension:

mult3 = [x for x in [1, 2, 3, 4, 5, 6, 7, 8, 9] if x % 3 == 0]

(or even as range(3,10,3)), but there are many other, more sophisticated use cases where you can’t use a list comprehension and a lambda function may be the shortest way to write something out.

• Returning a function from another function

  >>> def transform(n):
  ...     return lambda x: x + n
  ...
  >>> f = transform(3)
  >>> f(4)
  7
  

  This is often used to create function wrappers, such as Python’s decorators.

• Combining elements of an iterable sequence with reduce()

  >>> reduce(lambda a, b: '{}, {}'.format(a, b), [1, 2, 3, 4, 5, 6, 7, 8, 9])
  '1, 2, 3, 4, 5, 6, 7, 8, 9'
  

• Sorting by an alternate key

  >>> sorted([1, 2, 3, 4, 5, 6, 7, 8, 9], key=lambda x: abs(5-x))
  [5, 4, 6, 3, 7, 2, 8, 1, 9]
  

I use lambda functions on a regular basis. It took me a while to get used to them, but eventually I came to understand that they’re a very valuable part of the language.

lambda is just a fancy way of saying function. Other than its name, there is nothing obscure, intimidating or cryptic about it. When you read the following line, replace lambda by function in your mind:

>>> f = lambda x: x + 1
>>> f(3)
4

It just defines a function of x. Some other languages, like R, say it explicitly:

> f = function(x) { x + 1 }
> f(3)
4

You see? It’s one of the most natural things to do in programming.

The two-line summary:

1. Closures: Very useful. Learn them, use them, love them.
2. Python’s lambda keyword: unnecessary, occasionally useful. If you find yourself doing anything remotely complex with it, put it away and define a real function.

A lambda is part of a very important abstraction mechanism which deals with higher order functions. To get proper understanding of its value, please watch high quality lessons from Abelson and Sussman, and read the book SICP

These are relevant issues in modern software business, and becoming ever more popular.

lambdas are extremely useful in GUI programming. For example, lets say you’re creating a group of buttons and you want to use a single paramaterized callback rather than a unique callback per button. Lambda lets you accomplish that with ease:

for value in ["one","two","three"]:
    b = tk.Button(label=value, command=lambda arg=value: my_callback(arg))
    b.pack()

(Note: although this question is specifically asking about lambda, you can also use functools.partial to get the same type of result)

The alternative is to create a separate callback for each button which can lead to duplicated code.

I doubt lambda will go away. See Guido’s post about finally giving up trying to remove it. Also see an outline of the conflict.

You might check out this post for more of a history about the deal behind Python’s functional features: http://python-history.blogspot.com/2009/04/origins-of-pythons-functional-features.html

Curiously, the map, filter, and reduce functions that originally motivated the introduction of lambda and other functional features have to a large extent been superseded by list comprehensions and generator expressions. In fact, the reduce function was removed from list of builtin functions in Python 3.0. (However, it’s not necessary to send in complaints about the removal of lambda, map or filter: they are staying. :-)

My own two cents: Rarely is lambda worth it as far as clarity goes. Generally there is a more clear solution that doesn’t include lambda.

In Python, lambda is just a way of defining functions inline,

a = lambda x: x + 1
print a(1)

and..

def a(x): return x + 1
print a(1)

..are the exact same.

There is nothing you can do with lambda which you cannot do with a regular function—in Python functions are an object just like anything else, and lambdas simply define a function:

>>> a = lambda x: x + 1
>>> type(a)
<type 'function'>

I honestly think the lambda keyword is redundant in Python—I have never had the need to use them (or seen one used where a regular function, a list-comprehension or one of the many builtin functions could have been better used instead)

For a completely random example, from the article “Python’s lambda is broken!”:

To see how lambda is broken, try generating a list of functions fs=[f0,...,f9] where fi(n)=i+n. First attempt:

>>> fs = [(lambda n: i + n) for i in range(10)]
>>> fs[3](4)
13

I would argue, even if that did work, it’s horribly and “unpythonic”, the same functionality could be written in countless other ways, for example:

>>> n = 4
>>> [i + n for i in range(10)]
[4, 5, 6, 7, 8, 9, 10, 11, 12, 13]

Yes, it’s not the same, but I have never seen a cause where generating a group of lambda functions in a list has been required. It might make sense in other languages, but Python is not Haskell (or Lisp, or …)

Please note that we can use lambda and still achieve the desired results in this way :

>>> fs = [(lambda n,i=i: i + n) for i in range(10)]
>>> fs[3](4)
7

Edit:

There are a few cases where lambda is useful, for example it’s often convenient when connecting up signals in PyQt applications, like this:

w = PyQt4.QtGui.QLineEdit()
w.textChanged.connect(lambda event: dothing())

Just doing w.textChanged.connect(dothing) would call the dothing method with an extra event argument and cause an error. Using the lambda means we can tidily drop the argument without having to define a wrapping function.

I find lambda useful for a list of functions that do the same, but for different circumstances.

Like the Mozilla plural rules:

plural_rules = [
    lambda n: 'all',
    lambda n: 'singular' if n == 1 else 'plural',
    lambda n: 'singular' if 0 <= n <= 1 else 'plural',
    ...
]
# Call plural rule #1 with argument 4 to find out which sentence form to use.
plural_rule[1](4) # returns 'plural'

If you’d have to define a function for all of those you’d go mad by the end of it.
Also, it wouldn’t be nice with function names like plural_rule_1, plural_rule_2, etc. And you’d need to eval() it when you’re depending on a variable function id.

Pretty much anything you can do with lambda you can do better with either named functions or list and generator expressions.

Consequently, for the most part you should just one of those in basically any situation (except maybe for scratch code written in the interactive interpreter).

I’ve been using Python for a few years and I’ve never run in to a case where I’ve needed lambda. Really, as the tutorial states, it’s just for syntactic sugar.

I can’t speak to python’s particular implementation of lambda, but in general lambda functions are really handy. They’re a core technique (maybe even THE technique) of functional programming, and they’re also very useuful in object-oriented programs. For certain types of problems, they’re the best solution, so certainly shouldn’t be forgotten!

I suggest you read up on closures and the map function (that links to python docs, but it exists in nearly every language that supports functional constructs) to see why it’s useful.

Lambda function it’s a non-bureaucratic way to create a function.

That’s it. For example, let’s supose you have your main function and need to square values. Let’s see the traditional way and the lambda way to do this:

Traditional way:

def main():
...
...
y = square(some_number)
...
return something

def square(x):
    return x**2

The lambda way:

def main():
...
square = lambda x: x**2
y = square(some_number)
return something

See the difference?

Lambda functions go very well with lists, like lists comprehensions or map. In fact, list comprehension it’s a “pythonic” way to express yourself using lambda. Ex:

>>>a = [1,2,3,4]
>>>[x**2 for x in a]
[1,4,9,16]

Let’s see what each elements of the syntax means:

[] : “Give me a list”

x**2 : “using this new-born function”

for x in a: “into each element in a”

That’s convenient uh? Creating functions like this. Let’s rewrite it using lambda:

>>> square = lambda x: x**2
>>> [square(s) for x in a]
[1,4,9,16]

Now let’s use map, which is the same thing, but more language-neutral. Maps takes 2 arguments:

(i) one function

(ii) an iterable

And gives you a list where each element it’s the function applied to each element of the iterable.

So, using map we would have:

>>> a = [1,2,3,4]
>>> squared_list = map(lambda x: x**2, a)

If you master lambdas and mapping, you will have a great power to manipulate data and in a concise way. Lambda functions are neither obscure nor take away code clarity. Don’t confuse something hard with something new. Once you start using them, you will find it very clear.

One of the nice things about lambda that’s in my opinion understated is that it’s way of deferring an evaluation for simple forms till the value is needed. Let me explain.

Many library routines are implemented so that they allow certain parameters to be callables (of whom lambda is one). The idea is that the actual value will be computed only at the time when it’s going to be used (rather that when it’s called). An (contrived) example might help to illustrate the point. Suppose you have a routine which which was going to do log a given timestamp. You want the routine to use the current time minus 30 minutes. You’d call it like so

log_timestamp(datetime.datetime.now() - datetime.timedelta(minutes = 30))

Now suppose the actual function is going to be called only when a certain event occurs and you want the timestamp to be computed only at that time. You can do this like so

log_timestamp(lambda : datetime.datetime.now() - datetime.timedelta(minutes = 30))

Assuming the log_timestamp can handle callables like this, it will evaluate this when it needs it and you’ll get the timestamp at that time.

There are of course alternate ways to do this (using the operator module for example) but I hope I’ve conveyed the point.

Update: Here is a slightly more concrete real world example.

Update 2: I think this is an example of what is called a thunk.

As stated above, the lambda operator in Python defines an anonymous function, and in Python functions are closures. It is important not to confuse the concept of closures with the operator lambda, which is merely syntactic methadone for them.

When I started in Python a few years ago, I used lambdas a lot, thinking they were cool, along with list comprehensions. However, I wrote and have to maintain a big website written in Python, with on the order of several thousand function points. I’ve learnt from experience that lambdas might be OK to prototype things with, but offer nothing over inline functions (named closures) except for saving a few key-stokes, or sometimes not.

Basically this boils down to several points:

• it is easier to read software that is explicitly written using meaningful names. Anonymous closures by definition cannot have a meaningful name, as they have no name. This brevity seems, for some reason, to also infect lambda parameters, hence we often see examples like lambda x: x+1
• it is easier to reuse named closures, as they can be referred to by name more than once, when there is a name to refer to them by.
• it is easier to debug code that is using named closures instead of lambdas, because the name will appear in tracebacks, and around the error.

That’s enough reason to round them up and convert them to named closures. However, I hold two other grudges against anonymous closures.

The first grudge is simply that they are just another unnecessary keyword cluttering up the language.

The second grudge is deeper and on the paradigm level, i.e. I do not like that they promote a functional-programming style, because that style is less flexible than the message passing, object oriented or procedural styles, because the lambda calculus is not Turing-complete (luckily in Python, we can still break out of that restriction even inside a lambda). The reasons I feel lambdas promote this style are:

• There is an implicit return, i.e. they seem like they ‘should’ be functions.

• They are an alternative state-hiding mechanism to another, more explicit, more readable, more reusable and more general mechanism: methods.

I try hard to write lambda-free Python, and remove lambdas on sight. I think Python would be a slightly better language without lambdas, but that’s just my opinion.

Lambdas are actually very powerful constructs that stem from ideas in functional programming, and it is something that by no means will be easily revised, redefined or removed in the near future of Python. They help you write code that is more powerful as it allows you to pass functions as parameters, thus the idea of functions as first-class citizens.

Lambdas do tend to get confusing, but once a solid understanding is obtained, you can write clean elegant code like this:

squared = map(lambda x: x*x, [1, 2, 3, 4, 5])

The above line of code returns a list of the squares of the numbers in the list. Ofcourse, you could also do it like:

def square(x):
    return x*x

squared = map(square, [1, 2, 3, 4, 5])

It is obvious the former code is shorter, and this is especially true if you intend to use the map function (or any similar function that takes a function as a parameter) in only one place. This also makes the code more intuitive and elegant.

Also, as @David Zaslavsky mentioned in his answer, list comprehensions are not always the way to go especially if your list has to get values from some obscure mathematical way.

From a more practical standpoint, one of the biggest advantages of lambdas for me recently has been in GUI and event-driven programming. If you take a look at callbacks in Tkinter, all they take as arguments are the event that triggered them. E.g.

def define_bindings(widget):
    widget.bind("<Button-1>", do-something-cool)

def do-something-cool(event):
    #Your code to execute on the event trigger

Now what if you had some arguments to pass? Something as simple as passing 2 arguments to store the coordinates of a mouse-click. You can easily do it like this:

def main():
    # define widgets and other imp stuff
    x, y = None, None
    widget.bind("<Button-1>", lambda event: do-something-cool(x, y))

def do-something-cool(event, x, y):
    x = event.x
    y = event.y
    #Do other cool stuff

Now you can argue that this can be done using global variables, but do you really want to bang your head worrying about memory management and leakage especially if the global variable will just be used in one particular place? That would be just poor programming style.

In short, lambdas are awesome and should never be underestimated. Python lambdas are not the same as LISP lambdas though (which are more powerful), but you can really do a lot of magical stuff with them.

Lambdas are deeply linked to functional programming style in general. The idea that you can solve problems by applying a function to some data, and merging the results, is what google uses to implement most of its algorithms.

Programs written in functional programming style, are easily parallelized and hence are becoming more and more important with modern multi-core machines. So in short, NO you should not forget them.

First congrats that managed to figure out lambda. In my opinion this is really powerful construct to act with. The trend these days towards functional programming languages is surely an indicator that it neither should be avoided nor it will be redefined in the near future.

You just have to think a little bit different. I’m sure soon you will love it. But be careful if you deal only with python. Because the lambda is not a real closure, it is “broken” somehow: pythons lambda is broken

I’m just beginning Python and ran head first into Lambda- which took me a while to figure out.

Note that this isn’t a condemnation of anything. Everybody has a different set of things that don’t come easily.

Is lambda one of those ‘interesting’ language items that in real life should be forgotten?

No.

I’m sure there are some edge cases where it might be needed, but given the obscurity of it,

It’s not obscure. The past 2 teams I’ve worked on, everybody used this feature all the time.

the potential of it being redefined in future releases (my assumption based on the various definitions of it)

I’ve seen no serious proposals to redefine it in Python, beyond fixing the closure semantics a few years ago.

and the reduced coding clarity – should it be avoided?

It’s not less clear, if you’re using it right. On the contrary, having more language constructs available increases clarity.

This reminds me of overflowing (buffer overflow) of C types – pointing to the top variable and overloading to set the other field values…sort of a techie showmanship but maintenance coder nightmare..

Lambda is like buffer overflow? Wow. I can’t imagine how you’re using lambda if you think it’s a “maintenance nightmare”.

I use lambdas to avoid code duplication. It would make the function easily comprehensible Eg:

def a_func()
  ...
  if some_conditon:
     ...
     call_some_big_func(arg1, arg2, arg3, arg4...)
  else
     ...
     call_some_big_func(arg1, arg2, arg3, arg4...)

I replace that with a temp lambda

def a_func()
  ...
  call_big_f = lambda args_that_change: call_some_big_func(arg1, arg2, arg3, args_that_change)
  if some_conditon:
     ...
     call_big_f(argX)
  else
     ...
     call_big_f(argY)

I started reading David Mertz’s book today ‘Text Processing in Python.’ While he has a fairly terse description of Lambda’s the examples in the first chapter combined with the explanation in Appendix A made them jump off the page for me (finally) and all of a sudden I understood their value. That is not to say his explanation will work for you and I am still at the discovery stage so I will not attempt to add to these responses other than the following: I am new to Python I am new to OOP Lambdas were a struggle for me Now that I read Mertz, I think I get them and I see them as very useful as I think they allow a cleaner approach to programming.

He reproduces the Zen of Python, one line of which is Simple is better than complex. As a non-OOP programmer reading code with lambdas (and until last week list comprehensions) I have thought-This is simple?. I finally realized today that actually these features make the code much more readable, and understandable than the alternative-which is invariably a loop of some sort. I also realized that like financial statements-Python was not designed for the novice user, rather it is designed for the user that wants to get educated. I can’t believe how powerful this language is. When it dawned on me (finally) the purpose and value of lambdas I wanted to rip up about 30 programs and start over putting in lambdas where appropriate.

A useful case for using lambdas is to improve the readability of long list comprehensions. In this example loop_dic is short for clarity but imagine loop_dic being very long. If you would just use a plain value that includes i instead of the lambda version of that value you would get a NameError.

>>> lis = [{"name": "Peter"}, {"name": "Josef"}]

>>> loop_dic = lambda i: {"name": i["name"] + " Wallace" }
>>> new_lis = [loop_dic(i) for i in lis]

>>> new_lis
[{'name': 'Peter Wallace'}, {'name': 'Josef Wallace'}]

Instead of

>>> lis = [{"name": "Peter"}, {"name": "Josef"}]

>>> new_lis = [{"name": i["name"] + " Wallace"} for i in lis]

>>> new_lis
[{'name': 'Peter Wallace'}, {'name': 'Josef Wallace'}]

I can give you an example where I actually needed lambda serious. I’m making a graphical program, where the use right clicks on a file and assigns it one of three options. It turns out that in Tkinter (the GUI interfacing program I’m writing this in), when someone presses a button, it can’t be assigned to a command that takes in arguments. So if I chose one of the options and wanted the result of my choice to be:

print 'hi there'

Then no big deal. But what if I need my choice to have a particular detail. For example, if I choose choice A, it calls a function that takes in some argument that is dependent on the choice A, B or C, TKinter could not support this. Lamda was the only option to get around this actually…

I use it quite often, mainly as a null object or to partially bind parameters to a function.

Here are examples:

to implement null object pattern:

{
    DATA_PACKET: self.handle_data_packets
    NET_PACKET: self.handle_hardware_packets
}.get(packet_type, lambda x : None)(payload)

for parameter binding:

let say that I have the following API

def dump_hex(file, var)
    # some code
    pass

class X(object):
    #...
    def packet_received(data):
        # some kind of preprocessing
        self.callback(data)
    #...

Then, when I wan’t to quickly dump the recieved data to a file I do that:

dump_file = file('hex_dump.txt','w')
X.callback = lambda (x): dump_hex(dump_file, x)
...
dump_file.close()

I use lambda to create callbacks that include parameters. It’s cleaner writing a lambda in one line than to write a method to perform the same functionality.

For example:

import imported.module

def func():
    return lambda: imported.module.method("foo", "bar")

as opposed to:

import imported.module

def func():
    def cb():
        return imported.module.method("foo", "bar")
    return cb

I’m a python beginner, so to getter a clear idea of lambda I compared it with a ‘for’ loop; in terms of efficiency. Here’s the code (python 2.7) –

import time
start = time.time() # Measure the time taken for execution

def first():
    squares = map(lambda x: x**2, range(10))
    # ^ Lambda
    end = time.time()
    elapsed = end - start
    print elapsed + ' seconds'
    return elapsed # gives 0.0 seconds

def second():
    lst = []
    for i in range(10):
        lst.append(i**2)
    # ^ a 'for' loop
    end = time.time()
    elapsed = end - start
    print elapsed + ' seconds'
    return elapsed # gives 0.0019998550415 seconds.

print abs(second() - first()) # Gives 0.0019998550415 seconds!(duh)

Lambda is a procedure constructor. You can synthesize programs at run-time, although Python’s lambda is not very powerful. Note that few people understand that kind of programming.